This thesis addresses the problem of automatic 3D organoid classification within the field of 3D medical image classification, with application to monitoring the effects of endocrine disrupting chemicals (EDCs). A database of 3D volumes (full z-stacks) of organoids exposed to various EDCs was constructed, annotated according to three main phenotypes defined by morphology and cellular organization: cystic, compact, and cauliflower. The objective of this work was to develop and evaluate an end-to-end framework for the automatic classification of these phenotypes, specifically analyzing: the effectiveness of 3D deep learning architectures in distinguishing EDC-induced phenotypes; the comparison between transformer-based models (Swin-UNETR, SwinVit) and traditional 3D CNNs (ResNet, DenseNet)—motivated by the growing importance of transformers in computer vision; and the optimal trade-off between model complexity, computational efficiency, and predictive quality.